Upload
dinhnguyet
View
219
Download
0
Embed Size (px)
Citation preview
FORMAL APPROACHES TOFUNCTION IN GRAMMAR
Papers in honor of Eloise Jelinek
Edited By:
ANDREW CARNIEHEIDI HARLEYMARY WILLIE
University of Arizona
JOHN BENJAMINS PUBLISHING COMPANYAMSTERDAM/PHILADELPHIA
Table of Contents
List of Contributors
Acknowledgements
Introduction: Formalizing Functionalism
Andrew Carnie and Heidi Harley
Part I: The Pronominal Argument Hypothesis
On the Significance of Eloise Jelinek's Pronominal Argument Hypothesis
Ken Hale
Categories and pronominal arguments
Emmon Bach
Doubling by Agreement in Slave (Northern Athapaskan)
Keren Rice
Quasi Objects in St’át’imcets: On the semi independence of Agreement
and Case.
Henry Davis and Lisa Matthewson
Agreement, dislocation and partial configurationality
Mark Baker
Part II: Interfaces
Multiple multiple questions
Molly Diesing
Attitude evaluation in complex NPs
Lynn Nichols
Topic-Focus articulation and degrees of salience in the Prague
Dependency Treebank
Petr Sgall, Eva Hajic&ová and Eva Burác&ová
Word order and discourse genre in Tohono O'odham
Colleen M. Fitzgerald
The prosody of interrogative and focus constructions in Navajo
Joyce McDonough
Subject number, grammaticalization and transitivity in Cupeño
Jane H. Hill
Lexical Irregularity in OT: DOT vs. Variable Constraint Ranking
Diana Archangeli
Rapid perceptibility as a factor underlying universals of vowel inventories
Natasha Warner
Part III: Foundational Issues
Argument Hierarchies and the Mapping Principle
Eloise Jelinek and Andrew Carnie
Focus movement and the nature of uninterpretable features
Simin Karimi
Merge
Terry Langendoen
Phonotactics and Probablistic Ranking
Michael Hammond
Deconstructing Functionalist explanations for Linguistic Universals
Thomas G. Bever
Index
Contributors
Diana Archangeli, University of Arizona
Emmon Bach, University of Massachusetts, Amherst
Mark Baker, Rutgers University
Thomas Bever, University of Arizona
Eva Burác&ová, Charles University in Prague
Andrew Carnie, University of Arizona
Henry Davis, University of British Columbia
Molly Diesing, Cornell University
Colleen Fitzgerald, Texas Tech University
Eloise Jelinek, University of Arizona
Simin Karimi, University of Arizona
Ken Hale, Massachusetts Institute of Technology
Eva Hajic&ová, Charles University in Prague
Michael Hammond, University of Arizona
Jane Hill, University of Arizona
Terry Langendoen, University of Arizona
Lisa Matthewson, University of Massachusetts, Amherst
Joyce McDonough, University of Rochester
Lynn Nichols, University of California, Berkeley
Keren Rice, University of Toronto
Petr Sgall, Charles University in Prague
Natasha Warner, University of Arizona
Acknowledgments
A number of scholars helped in commenting on and refereeing the
papers in this volume. We would like to thank Mark Baker, Sonya Bird,
Noam Chomsky, Megan Crowhurst, Richard Demers, Sheila Dooley
Collberg, Chris Golston, Robert Franks, Paul Hagstrom, Daniel Hall-
Currie, Mike Hammond, Jason Haugen, Simin Karimi, Jaklin Kornfilt,
Terry Langendoen, Adrienne Lehrer, Alec Marantz, Joyce McDonough,
Martha McGinnis, Fritz Newmeyer, Massimo Piatelli-Palmarini, Norvin
Richards, Keren Rice, Leslie Saxon, Margeret Speas, Adam Ussishkin,
Andy Wedel, Natasha Warner, Mary Zampini for their help.
Introduction
Formalizing Functionalism
Andrew Carnie, Heidi Harley
University of Arizona
1. Formal vs. Functional
At least since Chomsky (1956, 1965), the field of linguistics has been
roughly divided into two camps concerning the philosophical,
methodological, and empirical domain of the discipline. Traditional
grammarians, following the early work of the Prague School of linguistics,
as well as linguists coming from approaches as diverse as computational
neural net theory and typological descriptive traditions, identify
themselves primarily as functional linguists. The term is meant to evoke
the idea that the driving force in linguistic study is a focus on the
(communicative) function of grammatical structure. The other approach,
which finds its origins in the American Structuralist school of linguistics
(such as the work by Bloomfield) but became solidified in Chomskyan
Generative Grammar and related approaches, is focused on the form and
structure of the language, rather than on its communicative (and other)
functions. This view is commonly called formalism.
The labels functionalist and formalist represent an unfortunate
oversimplification of very complex foundational issues. There is very little
agreement among linguists about which particular assumptions or
methodologies cause one to be a functionalist or formalist. There are a
number of important works on the question of defining the camps,
including Croft (1995) and Newmeyer (1998), and the collection of
articles in the volumes edited by Darnell, Moravcsik et al. (1999). We
cannot hope to do justice to the complexity of the issue here, but will offer
a brief synopsis of some of the leading lines of thought. Part of the
problem of defining the terms functional and formal lies in the fact that
among practitioners of functionalism there is little agreement as to what
characteristics a functional theory must have. On the side of formalist
theories (such as Minimalism, HPSG and LFG), there is a greater
agreement about the domain of the discipline, but still important
foundational disagreements about methodology and implementation. As
such, defining what constitutes a formal theory of grammar or a functional
one is notoriously difficult.
Essentially, the formal/functional distinction turns on one's
position along a number of dimensions. A theory's alignment as
functionalist or formalist seems to depend on how many of these
dimensions that one takes a more or less functional position on. In no
particular order these dimensions include (these dimensions are based on
work of Croft (1995) and Newmeyer (1998)):
(1) The role of structure in grammatical theory: Less reliance on
structure makes one more functionalist1.
(2) The role of arbitrariness in grammar: Formalists hold the strict
Saussurian view that language, including the grammar, is essentially
arbitrary; whereas the most radical of functionalist approaches only see
arbitrariness in the lexicon. Less radical functionalists (e.g. Cognitive
Grammar) view arbitrariness in the grammar and the lexicon to be of a
like-kind, that is: grammatical arbitrariness is essentially lexical
arbitrariness2.
3) The autonomy of syntax3: Many formalist theories hold that there
are a number of grammatical phenomena that allow formal
characterization without reference to their semantic or pragmatic function.
By contrast, all functionalist approaches view the importance of the
semantic or pragmatic function of grammatical phenomena to be so
overwhelming in shaping the grammatical form, that speaking of the
grammatical form in isolation is meaningless.
4) The diachronic/synchronic distinction: Formalist theories roughly
hold that the goal of linguistic theory is to characterize the grammatical
system of a speaker at a given moment in time, without reference to the
historical pressures that gave rise to that grammatical system.
Functionalists, hold that a full characterization of the grammatical system
is incomplete without an understanding of the historical events that gave
rise to it.
5) The competence/performance distinction4: Formalist grammarians
hold that there is core grammatical knowledge which can be characterized
independently of the production/comprehension system that realizes it.
Functionalist grammarians, on the whole, argue that the performance
system and the competence system are isomorphic.
6) A methodological issue over what constitutes "data" for linguistic
study: Formalist grammarians usually, although not uniformly, focus on
grammaticality judgements, typological comparison and data from
language acquisition. Functionalists often, but not always, exclude
grammaticality judgements, focusing instead on statistical corpus analysis.
They also include historical and sociological data which is regularly
excluded from formalist theories.
These broad distinctions among theoretical approaches have given
rise to an interesting split in the discipline with respect to particular
linguistic phenomena. Some phenomena are treated almost exclusively by
functionalist theoreticians, and others exclusively by formalists. For
example, with a few notable exceptions, the study of information
structure, and the discourse-syntax interface is primarily done in the
functionalist traditions. It is this distinction in terms of empirical coverage
that we refer to by the “‘Function” in the title of this book. In this volume
we gather together papers by well-known authors writing in formalist
(primarily Chomskyan) frameworks, about many of the phenomena
traditionally treated only in functionalist traditions.
2. Eloise Jelinek and Formal Functionalism
For the past 20 years, the work of Eloise Jelinek, while she self-identifies
as a die-hard formalist, has been a driving force in showing formalist
linguists how to treat material previously considered only within
functionalist traditions. Jelinek's work on lesser studied languages,
including Native American, Semitic and Australian languages, often
focussing on phenomena apparently controlled by external factors, such as
information structure, has shown other formal linguists how to approach
cases of interface between purely “grammatical” (i.e., syntactic,
morphological, or phonological) phenomena and the discourse, semantic,
phonetic or conceptual forces that drive them. Indeed, the most recent
version of the Chomskyan Principles and Parameters approach,
Minimalism (Chomsky 1995, 2001), explicitly claims that cross-linguistic
and language internal variation in the syntactic computational component
stems from the interface between that component and other systems such
as the conceptual/interpretive system or the auditory/production system.
Much of the credit for the change from purely syntactic modeling to
detailed consideration of interface conditions, should be attributed to
Eloise Jelinek (and, of course, her collaborators and colleagues who
investigate the same issues5).
Formal grammar is, for the most part6, structuralist. On a purely
surface level, many of (if not the majority of) the world's languages
exhibit syntactic phenomena that appear problematic for structure-based
accounts. Purely structural accounts, such as a phrase structure grammar
or a transformational grammar of the EST type, suggest that grammatical
and semantic information is read directly off of the structure of the
syntactic tree. For example, in the EST, the subject of an English sentence
is the daughter of the S node, and sister to the VP. Non-configurational7
languages, such as Warlpiri, allow a multitude of word orders (and by
hypothesis a multitude of structures). For such languages, a one to one
mapping between structure and grammatical/semantic function seems, at
least on the surface, to be severely misguided. This conclusion led many
functionalists and formalists (as in the LFG and RG traditions) to abandon
structural accounts of word order phenomena. Within the
structuralist/formalist traditions, many attempts were made to account for
these languages: Hale (1982) proposed that there was a configurationality
parameter, whereby non-configurational languages didn't use X-bar
structure, but used a non-configurational phrase structure rule instead:
S … X*… which allowed free ordering of elements. This type of
approach was challenged by the fact that even in non-configurational
languages there appear to be some phenomena which are sensitive to
hierarchical structure. Ross (1967) proposed that these languages had a
stylistic or purely phonological rule which took a fully specified
hierarchical structure and "scrambled8" the elements. Neither of these
approaches is terribly satisfying in light of the evidence that word order in
non-configurational languages, when examined in a closer light, appears
to be determined by discourse-level considerations, such as topic, focus or
narrative force. In traditional formal autonomous syntax, such "external"
factors are difficult to account for since they essentially involve access to
contextual (and thus presumably performance and other extralinguistic)
information. Jelinek's work, however, has shown that an autonomous
syntactic component does not necessarily have to operate in a vacuum.
Instead, by looking at the interfaces with such domains as discourse
information structure, we can distinguish between purely
syntactic/grammatical phenomena and those grammatical phenomena that
draw information from other modules.
Jelinek (1984), a seminal paper in this domain, proposed the
Pronominal Argument Hypothesis (PAH), which broke new ground in
providing an explanation of why non-configurational languages show
some sensitivity to structural hierarchical structural information, but
appear to resist an analysis of their surface word orders in terms of
structure. The PA hypothesis claims that languages are parameterized into
lexical argument languages (LA), which allow full NPs to be arguments of
a predicate, and pronominal argument languages (PA) which only allow
pronouns (or agreement which represents pronouns) in argument
positions. Full lexical NPs in PA languages are not in argument positions,
but are adjoined to the structure, as in left-dislocation structures in
English. The surface word order of lexical elements is determined by
discourse-configurational considerations (which may or may not also map
to particular structural positions in the tree, such as the specifiers of Topic
or Focus functional projections.) Phenomena that appear to be sensitive to
traditional hierarchical "syntactic" constraints make reference to the
pronominal arguments. Those that are sensitive to discourse refer to the
adjunct NPs.
Closely tied to this work on discourse-syntax interaction is
Jelinek's work on the mapping between the syntactic structure, semantic
notions such as specificity and definiteness, and grammatical notions
traditionally characterized in the functionalist literature9 in terms of
accessibility hierarchies (see, for example, Keenan (1976); Keenan and
Comrie (1977); Givon (1994) and many others). Because these notions
refer primarily to semantic and world-knowledge, they are not easily dealt
with in traditional autonomous-syntax models. Jelinek's work, both alone
and in collaboration with Molly Diesing, has proposed that such abstract
hierarchies correspond, in fact, to hierarchy imposed by the phrase
structure. Relational and semantic hierarchies are "mapped" off of the
syntactic structure. A number of such hierarchical effects, including
person, animacy, specificity, and topic/focus structure result from
mapping between specific positions in the syntactic tree and semantic
structure. Positions high in the syntactic tree correspond directly to items
that are high in the semantic hierarchy.
In pursuing these areas, Eloise Jelinek has shown that formal
Chomskyan grammar can, with minor modifications, provide explanatory
and interesting accounts of phenomena traditionally dealt with only by
functionalist theoreticians.
3. Formal Grammars of Function
The papers we have collected together in this volume continue in this
spirit. They either directly address questions that Jelinek has raised (e.g.,
the papers of Baker, Hale, Davis & Matthewson, and Bach) or pursue
other traditionally "functional" topics in the spirit of a formal approach
(e.g. Langendoen, Diesing, Rice, McDonough, Sgall et. al., etc.) The
papers address both syntactic/semantic/pragmatic questions and
phonological/phonetic ones, where the formal/functional divide also
exists.
Addressing to the relationship between Jelinek's Pronominal
Argument Hypothesis, the structural representation of grammatical
relations, and definiteness, Ken Hale presents a non-movement account of
the Navajo verb form and nominal incorporation, which he argues is a
necessary consequence of the PAH. Emmon Bach explores the
consequences of the PAH for composition, adopting a strict semantic
type/syntactic category mapping hypothesis. In particular, he suggests that
the way in which overt DP arguments are composed with verbal predicates
must differ radically in pronominal argument languages, since the verb-
word denotes a saturated proposition (or truth value), rather than an
argument-taking function. Bach concludes that the PAH provides a
framework in which one can characterize this type of parametric variation
as variation in the availability of V and VP as categories: these categories
are not available in PA languages. Also dealing with parametric variation
and the PAH, and perforce adopting a similar hypothesis about
composition of dislocated DP arguments, Mark Baker treats the semi-
configurational properties of Kinande, a Bantu language, relating them to
similar semi-configurational properties observed for Romance languages
by Alexiadou and Agnosopolou (1998). Baker relates these properties to
the presence or absence, and the obligatoriness or optionality, of subject
and object agreement in Kinnande. He proposes that a particular
formulation of the PAH can explain the behavior of agreeing and non-
agreeing DPs within a single language. Keren Rice examines the
interaction of agreement doubling with the famous y-/b-alternation in the
Northern Athabaskan language Slave. According to the PAH, pronominal
argument languages should always allow co-occurrence of overt DPs with
pronominal agreement (the pronominal arguments). Northern Athabaskan
languages, which in general do not allow doubling of agreement and overt
DP arguments, have been treated as non-PAH languages. Rice
demonstrates that in at least some contexts, Slave allows doubling of
agreement, and discusses the theoretical consequences of this result for the
PAH. In the last paper dealing explicitly with the PAH, Henry Davis and
Lisa Matthewson outline a treatment of some non-agreeing objects in
St'át'imcets (Straits Salish), which they argue are problematic for the PAH
approach to Salish. They demonstrate that the interaction of inherent
plurality of events and determiners on these objects can be naturally
treated as a scope phenomenon.
Turning to function more broadly construed as an interface issue,
Molly Diesing investigates multiple fronting in wh-questions in Yiddish.
This is a very current topic of investigation, and several formal accounts in
terms of w h-features have been recently proposed (Richards 1999,
Pesetsky 2000, among many others.) In the grammar of Yiddish, multiple
wh-fronting co-occurs with single-fronting and non-fronting constructions,
and Diesing shows that the multiply-fronted construction is strongly
conditioned by D(iscourse)-Linking, in the sense of Pesetsky 1987. Lynn
Nichols also deals with constraints on extraction, in the context of another
currently heated debate about the syntactic realization of Davidson’s event
argument. She proposes that the event argument itself can be relativized,
which produces a relative clause that has in the past been misanalyzed as a
clausal complement to ‘evaluative’ nominals like claim, fact, or theory.
Such nominals identify the speaker’s attitude toward the proposition;
Nichols shows that the relative-clause hypothesis can explain certain
morphological and syntactic facts in English, Chinese and Turkish. Her
paper is very much in the Jelinek tradition: she demonstrates the existence
of structural consequences of a factor (speaker’s evaluation of reliability)
which past formalists would have marginalized as not relevant to the
grammar. Eva Burác &ová, Eva Hajic &ová and Petr Sgall propose to
investigate the relationship between topic, focus and sentence structure in
the Functional Generative Description framework. They divide the
syntactic tree into Focus and Topic subtrees, much in the spirit of the
Mapping Hypothesis pioneered by Diesing and Jelinek.
In a third paper on focus constructions, looking at
phonology/discourse interactions, Joyce McDonough reports the results of
an important study of pitch excursions and focus in Navajo, and explores
the possibility that Jelinek's Pronominal Argument Hypothesis can provide
the basis for an explanation of the (lack of) intonational contours that
appear in Navajo focus constructions. Finally, Colleen Fitzgerald
investigates the three-way interface between discourse structure, syntactic
structure and rhythmic structure in Tohono O’odham, concluding in
particular that rhythmic constraints must also be taken into account in
determining surface word order in the language.
Morphology/syntax interface questions are addressed by Jane Hill
in her paper on the Cupeño verb and the appropriate functional projections
in which to situate Cupeño's valence-affecting affixes. Hill draws heavily
on Jelink's work on Yaqui (Hiaki, Yoeme) (Jelinek 1996), in which
Jelinek posits two valence-affecting functional heads: TransP and VoiceP.
Hill also discusses how a formal approach provides a grounding for the
historical and functional pressures that gave rise to the system. Diana
Archangeli’s paper investigates questions concerning the
morphology/phonology interface within Optimality Theory. She examines
the morpheme structure constraints operative in Tiv, concluding that a
theory which permits direct reference to morpheme classes by OT
constraints is empirically and theoretically superior to either a strictly
ranked grammar where irregularities are simply encoded as lexically
specified variation, or to a ‘co-phonology’ model where the phonological
idiosyncrasies of different morpheme classes can be associated with
different constraint rankings. In so doing she is able to account for
‘irregular’ variation in a systematic way that has previously escaped
formal treatment. Finally, examining the phonology/phonetics interface,
Natasha Warner investigates the relationship between the functional
dimension of segment perceptibility and typological frequency of
distinctive feature. She argues in support of the hypothesis that more
quickly perceived distinctions are more likely to show up as
grammaticized distinctive features in the grammars of languages. The
intuition is that the more quickly perceptible a feature is, the greater its
functional utility in speech, and that this pattern is reflected in the
typology of grammars.
In the final section, we present papers dealing with foundational
questions of grammatical structure and operations. The rule-based,
grammatical/ungrammatical generation systems of formalist doctrine
operate on the assumption that "extra-linguistic" factors, such as
frequency, should not play a part in the organization of the grammatical
system. Hammond, using the formal devices of Optimality theory,
explores the question of how we can incorporate these frequency effects
into a formal grammatical theory. In the spirit of the minimalist program,
Terry Langendoen presents a new proposal for the formal generation of
structured linguistic expressions, formally modifying Chomsky’s proposed
family of Merge operations, which are the heart of the generative engine
of modern formal linguistic theory. Langendoen’s conclusions about
Merge address the long-standing tension between the straightforwardly
functional limitations imposed by computational capacity and the formal
expressiveness needed to generate the recursive structures of natural
language. Also working within the minimalist program, Simin Karimi
addresses the foundational issue of the nature of the features whose
checking drives movement, in particular examining the theoretical status
of so-called ‘uninterpretable’ features, contrasted with interpretable
features which drive the movement of wh-phrases. Discussing two types
of focus construction in Persian, only one of which seems to involve overt
movement, she argues that focus is indeed an instance of the operation
Move, motivated by the need to check a syntactic feature. Finally, Tom
Bever asks the question that formalists avoid and functionalists embrace:
given that the formal grammatical systems are the way (formalists say)
they are, why are they that way? Bever offers some general constraints on
functionalist answers to this question, and broadens the scope of what is
usually considered a functional explanation. He considers this alternative
version of function-as-explanation from three perspectives: language
perception, language acquisition, and the biological sources of linguistic
ability.
Sadly, Ken Hale passed away in October, 2001 before this volume was
completed. He was a driving force behind this volume, encouraging us to
put it together and was a great fan of the work of Eloise Jelinek.
1 Although there are many structuralist functionalist theories, such as Dik's
Functionalism or Van Valin's Role and Reference Grammar. See Noonan
(1999) for discussion of the role of structuralism in functionalist theories.
2 See Langacker (1987) on the list-rule fallacy.
3 See Newmeyer (1998) for a discussion of the various usages of this term
(meaning variously: autonomy of the grammatical system, autonomy of
the competence system, or autonomy of the syntactic system).
4 This question is intimately connected to the autonomy question
mentioned in footnote 3 above and the data issue listed in number (6).
5 such as Ken Hale, Ellen Prince, Molly Diesing, to name only a few.
6 We exclude Categorial Grammar and LFG, since these theories, although
formal, don't take a structuralist approach to the phenomena discussed
here.
7 The term "non-configurational" has been variously used to describe free
word order or scrambling languages, as well as languages that have fixed
word order but may lack the hierarchical effects associated with having a
VP (such as Japanese, Navajo and VSO languages, see Speas (1990) for
discussion). We are using the term only in the former (free word order)
usage here. However, Jelinek's approach extends to the latter usage as
well.
8 Ross's notion of scrambling and the current conception of the
phenomenon are very different, see below and Karimi (this volume) for
discussion.
9 Recently, a number of papers in Optimality Theory have also attempted
to deal with these questions, see Aissen (1999); Bresnan, Dingare et al.
(2001) for more details.